Abstract: One effective way to accommodate the large number of small-scale distributed wind power generation,photovoltaic power generation and energy storage units is to aggregate them by energy aggregators,and to participate in virtual power plants（VPPs）for optimal dispatch. However,it is usually difficult for VPP dispatch center to build detailed models and make accurate forecasts for the overall output characteristics of the aggregators,which brings challenges to the traditional centralized dispatch of VPPs. An interactive dispatch model based on deep reinforcement learning（DRL）is presented for VPPs containing distributed wind generation units,distributed photovoltaic generation units and distributed energy storage units. Through the online information interaction with the aggregators,the VPP dispatch center gradually learns the aggregate output of various units in VPP and the purchase and sale decision of VPP for the large power grid. The dispatch model is solved by using deep deterministic policy gradient（DDPG）algorithm. Examples based on real data are presented to demonstrate the effectiveness of proposed method. By comparing the results of our method with those of the traditional centralized dispatch obtained by CBC linear programming solver,it is shown that the proposed method is helpful in increasing total benefits of VPP,especially in improving the rate of renewable power utilization.